In Time Difference of Arrival (TDOA)-based sound source localization methods, the estimation accuracy of TDOA is constrained and limited by the sampling frequency. A low sampling frequency can lead to significant errors between estimated and true time delays, further resulting in complete deviation between estimated and actual positions of distant sound sources. Additionally, using a single microphone array for sound source localization results in significant errors, failing to meet practical application requirements.To address these issues, an improved second-order cross-correlation algorithm based on cubic spline interpolation for TDOA estimation is proposed. Cubic spline interpolation is employed after the first cross-correlation to increase the sampling frequency, followed by the second cross-correlation to obtain TDOA. Simulation results under conditions of 10-fold interpolation demonstrate that this method reduces the TDOA error from 7.6% to 0.6%.Subsequently, a sound source localization method based on data fusion of multiple microphone arrays is proposed. The localization results of multiple microphone arrays are fused to obtain the final position of the sound source. Taking a quaternion cross array as an example, 10 points in space are selected to compare the far-field localization performance of single, dual, and quaternion microphone arrays. The results indicate that the localization error of a single microphone array can exceed 1 meter when the sound source is distant, while the localization errors of dual and quaternion microphone arrays are generally within 0.3 meters and 0.2 meters, respectively, significantly improving localization accuracy.。